Self-cleaning bird repelling method and system for navigation mark and storage medium

Abstract

The invention relates to the technical field of power system safety and environmental protection, in particular to a self-cleaning bird repelling method and system of a navigation mark and a storage medium. According to the invention, deep learning, image recognition and automatic control technologies are combined, and different bird repelling and cleaning modes are automatically selected through detection of birds and analysis of bird droppings coverage conditions. The system does not need manual intervention, adopts advanced data fusion and time synchronization technologies, automatically identifies bird types and bird droppings conditions, formulates corresponding coping strategies, and avoids excessive cleaning and bird repelling and saves resources by accurately selecting bird repelling and cleaning modes. The cleaning strength is adjusted in real time according to the coverage rate of bird droppings, the navigation mark facility is kept clean, bird inhabitation is effectively reduced, the maintenance efficiency of the navigation mark facility is improved, damage and pollution of birds to equipment are avoided, and meanwhile the requirement for manual intervention is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of aid - to - navigation maintenance management, and particularly relates to a self - cleaning bird - repelling method, system and storage medium for aids - to - navigation. Background Art

[0002] The activities of birds have also increased significantly. The aid - to - navigation, known as the "eye of the ship", has become a temporary resting place and habitat for birds on the river surface. A large amount of bird droppings seriously pollute the aid - to - navigation itself. For example, bird droppings may block the light or cover the sensors, affecting the normal operation of the equipment and reducing the navigation function of the aid - to - navigation. To solve the problem of bird - dropping pollution, the self - cleaning technology on the aid - to - navigation needs to be able to automatically clean after birds inhabit, avoiding manual intervention. In the prior art, the automated cleaning technology mainly relies on mechanical cleaning, spray cleaning or self - cleaning coating and other technologies.

[0003] In the prior art, there is a lack of a cleaning system. The main purpose of bird - repelling is to reduce the accumulation of bird droppings in the use scenario. And the monitoring system has used various types, and the fusion recognition of multiple signals requires a high computing level for the central processing unit. There is a lack of monitoring of the system operation itself. At the same time, laser and strong - sound bird - repelling have a greater impact on the surrounding environment, may cause unnecessary harm to birds, and increase the maintenance and operation costs. Summary of the Invention

[0004] In view of the above - mentioned defects of the prior art, the present application provides a self - cleaning bird - repelling method, system and storage medium for aids - to - navigation, which combines deep learning, image recognition and automation control technologies. By detecting birds and analyzing the coverage of bird droppings, different bird - repelling and cleaning modes are automatically selected to keep the aid - to - navigation facilities clean, effectively reduce bird inhabitation, improve the maintenance efficiency of the aid - to - navigation facilities, avoid damage and pollution of the equipment by birds, and reduce the need for manual intervention at the same time.

[0005] In the first aspect, a self - cleaning bird - repelling method for aids - to - navigation, the method includes:

[0006] S1: Collect image data of a preset monitoring range and perform data annotation to obtain a target data set; wherein, the target data set includes information on different types of birds and environmental information;

[0007] S2: Input the target data set into a pre - trained deep - learning model to identify the existing birds and calculate the corresponding bird information, where the bird information includes the type and location of the birds;

[0008] S3: Input the target data with birds into an image recognition model to detect bird droppings and obtain the fecal coverage rate;

[0009] S4: Select to activate the first bird repelling mode or the second bird repelling mode according to the bird information calculated in S2; meanwhile, select different cleaning modes according to the fecal coverage rate calculated in S3.

[0010] The method proposed in this application collects environmental image data within a preset monitoring range, performs annotation to form a target data set containing different bird species and environmental information. Input the annotated target data set into a pre-trained deep learning model to identify the birds in the image and calculate their corresponding information. Analyze the distribution of bird droppings in the bird inhabiting area through an image recognition model, and calculate the fecal coverage rate. According to the bird information and the fecal coverage rate, select appropriate bird repelling modes and cleaning modes. It combines deep learning, image recognition and automation control technologies to improve the maintenance efficiency of navigation mark facilities, avoid damage and pollution of equipment by birds, and at the same time reduce the need for manual intervention.

[0011] Preferably, when performing data annotation in step S1, it further includes:

[0012] Use a public data set to perform data annotation on the collected image data; among them, the public data set is at least the Birdsnap data set or the CUB-200 data set; the data annotation includes existing bird annotation and possible bird annotation.

[0013] Preferably, before step S2, it further includes:

[0014] Preprocess the collected image data, and the preprocessing includes denoising, adjusting brightness and contrast for the target data set;

[0015] In the image data obtained after preprocessing, the bird droppings are grayish-white; the navigation marks are black, red or yellow.

[0016] Preferably, step S3 further includes:

[0017] S31: Convert the image from BGR to HSV or Lab color space;

[0018] S32: Set a threshold according to the color characteristics of the bird droppings to extract the fecal area;

[0019] After extracting the fecal area, use dilation and erosion operations to remove noise and fill small holes, use the findContours function to detect the extracted bird droppings area, and use the Gaussian blur method for smoothing and denoising to optimize the segmentation effect.

[0020] Preferably, step S3 further includes:

[0021] After segmenting the area covered by the feces, calculate the proportion of the feces in the entire image occupying the navigation mark area, and the formula is:

[0022] Fecal coverage rate = (number of pixels in the fecal area / number of pixels in the overall base area) × 100%;

[0023] It is divided into grade A, grade B and grade C according to the fecal coverage rate, and corresponding cleaning modes are set according to different grades;

[0024] Among them, grade A means the coverage rate is low and the cleaning work is completed; grade B means the coverage rate is high and cleaning needs to be started; grade C means the coverage rate is very high and manual cleaning is required.

[0025] Preferably, in step S4, according to the bird information calculated in S2, selecting to start the first bird repelling mode or the second bird repelling mode further includes:

[0026] When the bird position is above the beacon or in the surrounding area, start the first cleaning mode; the first cleaning mode is to turn on the variable frequency ultrasonic device, and the ultrasonic power amplifier frequency continuously changes between 20Khz - 70Khz;

[0027] When the bird position is on the base below the beacon, start the second cleaning mode, and the second cleaning mode is to trigger the high - pressure water flow and air flow device so that the impact range of the water flow and air flow can cover the entire beacon base.

[0028] Preferably, in step S4, according to the fecal coverage rate calculated in S3, selecting different cleaning modes further includes:

[0029] During the cleaning process of the system, the overall operation time is divided into T working time slots, and each unit t time slot is a cleaning cycle. The present invention designs a method for the adaptive cleaning of beacons based on deep reinforcement learning. Each beacon is regarded as an agent to interact with the environment and make decisions on adaptive cleaning actions. The beacon agent will cause the system to transition to a new state by performing different cleaning actions. In the next time slot, the agent makes a more adaptive cleaning action by analyzing the new state. Finally, by continuously interacting with the environment, the reward obtained by the agent will tend to be stable. At this moment, the agent has learned to execute the adaptive cleaning action and will make the optimal cleaning action in each time slot. First, the three key factors in this algorithm, namely the state space, action space, and reward, are defined as follows:

[0030] According to the bird repelling and cleaning process, the state space is defined as the fecal coverage rate C and the number of cleaning operations n l . Then the state at the t - th time slot is represented as s(t) = [C(t), N(t)];

[0031] Among them, C(t) = [c 1 (t),..., c l (t),..., c L(t)] represents the fecal coverage rate matrix of the navigation buoy at time slot t; N(t) = [n 1 (t),..., n l (t),..., n L (t)] represents the cleaning times of the navigation buoy at each time slot;

[0032] The action at each time slot is defined as the scheduling strategy for cleaning the navigation buoy, denoted as a(t) = [K(t), F(t)];

[0033] where K(t) = [k 1 (t),..., k l (t),..., k L (t)] represents the cleaning scheduling execution matrix of the navigation buoy nodes at each time slot, where k l (t) = {0, 1, 2} represents the scheduling status of the cleaning function at each time slot, 0 means no cleaning is required at this moment, 1 means cleaning needs to be started at this time slot, and 2 means manual cleaning is required; F(t) = [f 1 (t),..., f 1 (t),..., f L (t)] represents the number of water sprays required for each cleaning of each navigation buoy, where f 1 (t) = {0, 1, 2, 3}.

[0034] The reward at time slot t is defined as a linear function of the overall utility of the system observed at the end of time slot t, denoted as r(t) = -λ∑ l∈L c l (t), where λ is a constant, representing the opposite of the bird feces coverage area on the navigation buoy; so the sum of future rewards can be defined as:

[0035]

[0036] where T and 0 < γ < 1 are the last step of each cycle and the discount factor representing the influence of future rewards respectively; The present invention uses the DQN framework to optimize the strategy, where DQN uses an artificial neural network Q(s, a; θ), called the prediction network, as a function approximator to estimate the action value function, and θ is the weight of the neural network;

[0037] The input of the prediction network is the state s, and the corresponding values of all possible actions are generated as the output;

[0038] The target network is the neural network Q'(s, a; θ'), used to estimate the target value; The target network has the same structure as the prediction network; The weight θ' of the target network is copied from θ at every fixed number of iterations n instead of every training cycle;

[0039] According to the actions taken by the agent, the lighthouse cleaning environment will transition to a new state s(t+1); the agent collects an immediate reward r(t) from the environment and updates the weights of the neural network in the algorithm by minimizing the loss function L(θ) at each step:

[0040] L(θ) = [(y j -Q(s j , a j ; θ)) 2 ; y j = r j + γmax a' Q'(s j+1 , a'; θ').

[0041] Finally, the neural network will converge to a state where the reward stabilizes. At this time, the agent (lighthouse) has learned to adaptively adjust the cleaning actions to fit the dynamic cleaning environment.

[0042] Preferably, the self-cleaning bird repelling method of the lighthouse further includes: adopting orthogonal frequency division multiple access technology as the access scheme for different communication links; specifically:

[0043] Define a set of lake lighthouse collections as where l represents the first lighthouse; each lighthouse is equipped with a terminal device monitoring function, which can monitor the various system functions on the lighthouse and transmit the monitoring status of each part to the ground central processor through wireless signals. Then, when the ground central processor receives the signal sent by the lighthouse l, it is expressed as:

[0044]

[0045] The transmission rate from the lighthouse l to the central processing system is expressed as:

[0046] R l，u = B l，u log 2 (1 + γ l，u );

[0047] where p l represents the signal transmission power of the lighthouse l, x l represents the transmitted signal of the lighthouse l, h l，u represents the channel gain between the lighthouse l and the ground station, where the noise is the power of the noise; represents the transmission signal-to-noise ratio, B l，u represents the transmission bandwidth, and the faster the transmission rate, the better the transmission performance of the system.

[0048] Second aspect, a self-cleaning bird repelling system for a lighthouse, the system includes:

[0049] A data acquisition module for collecting image data within a preset monitoring range, performing data annotation, and obtaining a target data set;

[0050] A first processing module that inputs the target data set into a pre-trained deep learning model to identify existing birds and calculates the corresponding bird information;

[0051] A second processing module that inputs the target data with existing birds into an image recognition model for bird droppings detection to obtain the droppings coverage rate;

[0052] A first control module that selects to activate the first bird repelling mode or the second bird repelling mode according to the calculated bird information; a second control module that selects to activate the first cleaning mode or the second cleaning mode according to the calculated droppings coverage rate;

[0053] And a transmission module that uses orthogonal frequency division multiple access technology as the communication link between the ground central processing system and the navigation aid.

[0054] The system proposed in this application, with an efficient, precise, and intelligent operation mode, can accurately monitor the behavior of birds through advanced image recognition technology, deep learning models, automated control, and efficient communication technology, automatically select appropriate bird repelling and cleaning strategies, maximize work efficiency, reduce operating costs, and reduce manual intervention.

[0055] In a third aspect, a computer program is stored in the storage medium, where the computer program is configured to execute the self-cleaning bird repelling method of a navigation aid described in any one of the above when running.

[0056] The self-cleaning bird repelling method, system, and storage medium of a navigation aid proposed in this application, with an efficient, precise, and intelligent operation mode, can accurately monitor the behavior of birds through advanced image recognition technology, deep learning models, automated control, and efficient communication technology, automatically select appropriate bird repelling and cleaning strategies, maximize work efficiency, reduce operating costs, and reduce manual intervention.

[0057] Compared with the prior art, the beneficial effects of this application are as follows:

[0058] 1. Automatically identify birds and bird droppings, and intelligently select appropriate bird repelling and cleaning modes.

[0059] 2. Utilize deep learning and image recognition technology to accurately identify birds and bird droppings.

[0060] 3. Automatically adjust the bird repelling and cleaning modes according to real-time data.

[0061] 4. Have a stable and fast communication link.

[0062] 5. Adapt to remote monitoring and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a logic flowchart of a self - cleaning bird - repelling method for a navigation mark shown in an embodiment of the present application.

[0064] Figure 2 is a logic flowchart of a self - cleaning bird - repelling system for a navigation mark shown in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions will be clearly and completely described below in conjunction with the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0066] Embodiment 1: The present application proposes a self - cleaning bird - repelling method, system, and storage medium for a navigation mark, which is different from the existing bird - repelling methods. The cleaning system of the present application includes two parts: water cleaning and super - hydrophobic coating application. Necessary monitoring means are proposed, which can complete identification and monitoring more efficiently and reduce economic costs. A signal back - transmission system is added to monitor whether the overall bird - repelling system at the remote end is operating normally in real time. If there is a fault, a report can be made at any time. At the same time, it is proposed to use a water gun to spray birds and ultrasonic waves outside the human hearing range to repel birds to reduce the impact.

[0067] As shown in the Figure 1 accompanying drawings, a self - cleaning bird - repelling method for a navigation mark is as follows:

[0068] S1: Collect image data for a preset monitoring range and perform data annotation to obtain a target data set; wherein, the target data set includes information on different types of birds and environmental information;

[0069] S2: Input the target data set into a pre - trained deep - learning model to identify the existing birds and calculate the corresponding bird information, where the bird information includes the type and location of the birds;

[0070] S3: Input the target data with birds into an image recognition model for bird droppings detection to obtain the fecal coverage rate;

[0071] S4: According to the bird information calculated in S2, select to activate the first bird - repelling mode or the second bird - repelling mode; at the same time, according to the fecal coverage rate calculated in S3, select different cleaning modes.

[0072] The method proposed in this application collects environmental image data within a preset monitoring range, performs annotation, and forms a target data set containing different bird species and environmental information. The annotated target data set is input into a pre-trained deep learning model to identify the birds in the image and calculate their corresponding information. The distribution of bird droppings in the bird habitat area is analyzed through an image recognition model, and the coverage rate of the droppings is calculated. According to the bird information and the bird dropping coverage rate, appropriate bird repelling modes and cleaning modes are selected. By combining deep learning, image recognition, and automation control technologies, the maintenance efficiency of navigation mark facilities is improved, damage and pollution of the equipment by birds are avoided, and at the same time, the need for manual intervention is reduced.

[0073] Preferably, in step S1 for data annotation, it further includes:

[0074] Using a public data set to perform data annotation on the collected image data; wherein, the public data set is at least the Birdsnap data set or the CUB-200 data set; the data annotation includes existing bird annotation and possible bird annotation. Among them, Birdsnap contains an image data set of more than 500 bird species. The images of each bird species are annotated with specific species and include rich features (such as the shape, color, posture, etc. of the bird) for bird recognition and classification tasks. The CUB-200 data set is a bird image data set containing images of 200 different bird species, with at least 30 images for each bird species. In addition, it also includes detailed annotations for each image, such as the bounding box, keypoints, and species label of the bird.

[0075] Preferably, before step S2, it further includes:

[0076] Preprocessing the collected image data, and the preprocessing includes denoising, adjusting brightness and contrast for the target data set;

[0077] Preferably, for denoising the target data set, Gaussian filtering can be optionally used to perform convolution on the image using a Gaussian kernel function to make the image smoother and thus remove noise. However, it is not limited to this.

[0078] Preferably, the brightness is adjusted by increasing or decreasing the value of each pixel in the image to change the overall brightness of the image.

[0079] Preferably, the contrast is adjusted by modifying the alpha value, and the larger the alpha value, the higher the contrast.

[0080] In the image data obtained after preprocessing, the bird droppings are grayish-white; the navigation marks are black, red, or yellow.

[0081] Preferably, step S3 further includes:

[0082] S31: Convert the image from BGR to the HSV or Lab color space;

[0083] Among them, HSV (hue, saturation, value): is more suitable for color segmentation. Lab (lightness, green - red channel, blue - yellow channel): is more robust to illumination changes. Converting the image from BGR to the HSV or Lab color space can help with better color segmentation.

[0084] S32: Set a threshold according to the color characteristics of the bird droppings to extract the droppings area;

[0085] Among them, the color of bird droppings is usually grayish - white. Therefore, in the HSV color space, the H value of grayish - white is small, the S value is low, and the V value is high. Extract the bird droppings area by setting appropriate HSV thresholds. An exemplary operation is as follows:

[0086] # Set the HSV threshold range to extract the grayish - white area (according to the color of bird droppings);

[0087] lower_gray = np.array([0, 0, 180]) # Low saturation, high value;

[0088] upper_gray = np.array([180, 50, 255]) # The maximum H value is 180, suitable for grayish - white;

[0089] # Use the inRange function to extract the bird droppings area;

[0090] mask_gray = cv2.inRange(hsv_image, lower_gray, upper_gray);

[0091] # Display the segmented area;

[0092] cv2.imshow('Gray Region Mask', mask_gray);

[0093] cv2.waitKey(0);

[0094] cv2.destroyAllWindows();

[0095] In the above, the inRange function extracts the grayish - white area in the image according to the set color range, and generates a binary mask for the qualified part. This mask can be used for further image processing later.

[0096] S33: After extracting the fecal area, dilation and erosion operations are used to remove noise and fill small holes. The findContours function is used to detect the extracted bird feces area, and the Gaussian blur method is used for smoothing and denoising to optimize the segmentation effect.

[0097] Since some noise (small white dots or small areas) are often obtained after extracting the fecal area, preferably, dilation and erosion operations are performed to further remove noise and fill small holes to improve the segmentation effect. The exemplary operations are as follows:

[0098] # Define the structural element (kernel), a 5x5 square;

[0099] kernel = np.ones((5, 5), np.uint8);

[0100] # Erosion operation (removing small noise points);

[0101] eroded_mask = cv2.erode(mask_gray, kernel, iterations = 1);

[0102] # Dilation operation (filling small holes);

[0103] dilated_mask = cv2.dilate(eroded_mask, kernel, iterations = 2);

[0104] # Display the mask after dilation and erosion;

[0105] cv2.imshow('Processed Mask', dilated_mask);

[0106] cv2.waitKey(0);

[0107] cv2.destroyAllWindows();

[0108] Finally, the findContours function is used to detect the contours in the image, and then the area of the bird feces is extracted. findContours returns a list of contours.

[0109] Preferably, step S3 further includes:

[0110] After segmenting the area covered by feces, calculate the proportion of feces in the entire image occupying the beacon area. The formula is:

[0111] Fecal coverage rate = (number of pixels in the fecal area / number of pixels in the overall base area) × 100%;

[0112] It is divided into grade A, grade B and grade C according to the fecal coverage rate, and corresponding cleaning modes are set according to different grades;

[0113] Among them, grade A has a low coverage rate and the cleaning work is completed; grade B has a high coverage rate and cleaning needs to be started; grade C has a very high coverage rate and manual cleaning is required.

[0114] By converting the color space, setting thresholds, morphological processing and Gaussian blur, the bird feces area can be accurately identified, noise interference can be reduced, the segmentation effect can be ensured to be more accurate, the fecal coverage rate can be calculated and graded, the cleaning start time can be determined, the cleaning operation efficiency can be optimized, over-cleaning or omission can be avoided. At the same time, according to different coverage rates, different cleaning methods are selected, which improves the efficiency and flexibility of the overall cleaning management, automatically selects the cleaning intensity and mode, avoids resource waste, and ensures the cleanliness of the environment. Through these optimization steps, the system can more intelligently identify and handle bird feces problems, thereby improving the overall management and operation efficiency.

[0115] Preferably, in step S4, according to the bird information calculated in S2, the first bird repelling mode or the second bird repelling mode is selected to be started, and it further includes:

[0116] When the bird's position is above the beacon or in the surrounding area, the first cleaning mode is started; the first cleaning mode is to turn on the variable-frequency ultrasonic device, and the ultrasonic power amplifier frequency continuously changes between 20Khz - 70Khz; the continuous change of the frequency can increase the repelling effect of the ultrasonic wave because the frequency conversion will disrupt the adaptive response of the birds, making it impossible for them to adapt or resist.

[0117] When the bird's position is on the lower base of the beacon, the second cleaning mode is started, and the second cleaning mode is to trigger the high-pressure water flow and air flow device so that the impact range of the water flow and air flow can cover the entire beacon base.

[0118] Preferably, in step S4, according to the fecal coverage rate calculated in S3, different cleaning modes are selected, and it further includes:

[0119] During the cleaning process of the system, the overall operation time is divided into T working time slots, and each unit t time slot is a cleaning cycle. The present invention designs a method for adaptive cleaning of beacons based on deep reinforcement learning, taking each beacon as an agent to interact with the environment and make adaptive cleaning action decisions. The beacon agent will cause the system to transition to a new state by performing different cleaning actions. In the next time slot, the agent makes a more adaptive cleaning action by analyzing the new state. Finally, through continuous interaction with the environment and training of the decision-making agent, the obtained reward will tend to a certain stable trend. At this moment, the agent has learned to execute the adaptive cleaning action and will make the optimal cleaning action in each time slot. First, the three key factors in this algorithm, namely the state space, action space and reward, are defined as follows:

[0120] According to the bird repelling and cleaning process, the state space is defined as the fecal coverage rate C and the number of cleaning operations n l . Then the state at the t-th time slot is represented as s(t) = [C(t), N(t)];

[0121] where C(t) = [c 1 (t),..., c l (t),..., c L (t)], representing the fecal coverage rate matrix of the navigation mark at time slot t; N(t) = [n 1 (t),..., n l (t),..., n L (t)] represents the number of cleaning times of the navigation mark for each time slot;

[0122] The action for each time slot is defined as the scheduling strategy for cleaning the navigation mark, represented as a(t) = [K(t), F(t)];

[0123] where K(t) = [k 1 (t),..., k l (t),..., k L (t)], representing the cleaning scheduling execution matrix of the navigation mark nodes for each time slot, where k l (t) = {0, 1, 2} represents the scheduling state of the cleaning function for each time slot, 0 means no cleaning is required at this moment, 1 means cleaning needs to be started in this time slot, and 2 means manual cleaning is required; F(t) = [f 1 (t),..., f 1 (t),... f L (t)], representing the number of times of spraying water required for each cleaning of each navigation mark, where f 1 (t) = {0, 1, 2, 3}.

[0124] The reward at time slot t is defined as a linear function of the overall utility of the system observed at the end of time slot t, represented as r(t) = -λΣ l∈L c l (t), where λ is a constant, representing the opposite of the area covered by bird feces on the navigation mark; Therefore, the sum of future rewards can be defined as:

[0125]

[0126] where T and 0 < γ < 1 are the last step of each cycle and the discount factor representing the influence of future rewards respectively; The present invention uses the DQN framework to optimize the strategy, where DQN uses an artificial neural network Q(s, a; θ), called the prediction network, as a function approximator to estimate the action value function, and θ is the weight of the neural network;

[0127] The input of the prediction network is the state s, and the corresponding values of all possible actions are generated as the output;

[0128] The target network is a neural network Q'(s, a; θ'), which is used to estimate the target value; the target network has the same structure as the prediction network; the weights θ' of the target network are copied from θ at every fixed number of iterations n instead of every training cycle;

[0129] According to the actions taken by the agent, the buoy cleaning environment will transition to a new state s(t+1); the agent collects the immediate reward r(t) from the environment and updates the weights of the neural network in the algorithm by minimizing the loss function L(θ) at each step:

[0130] L(θ) = [(y j -Q(s j ,a j ;θ)) 2 ;y j =r j +γmax a' Q'(s j+1 ,a'; θ').

[0131] Finally, the neural network will converge to a state where the reward tends to be stable, and at this time, the agent (buoy) has learned to adaptively adjust the cleaning actions to fit the dynamic cleaning environment.

[0132] Through deep reinforcement learning and dynamic decision-making mechanisms, the cleaning strategy is flexibly adjusted according to the specific situation of each buoy, maximizing the cleaning efficiency and accuracy, avoiding over-cleaning or omission, saving cleaning resources, improving the overall operation efficiency of the system, reducing the interference of birds to the buoys, and at the same time keeping the buoys clean. Based on the adaptive mechanism of deep reinforcement learning, the system continuously learns and improves, can make more accurate and efficient cleaning decisions in a more complex environment, combines the bird position, bird droppings coverage rate with the deep reinforcement learning algorithm, realizes highly intelligent adaptive cleaning control of the buoys, adjusts according to the real-time environment, optimizes the decision-making through continuous learning, ensures that the cleaning work is always carried out efficiently and accurately, and at the same time optimizes the resource use.

[0133] Preferably, the self-cleaning bird repelling method for a buoy further includes: adopting orthogonal frequency division multiple access technology as the access scheme for different communication links; specifically:

[0134] Define a set of lake surface buoy collections as Among them, l represents the first navigation mark; each navigation mark is equipped with a terminal device monitoring function, which can monitor the various system functions on the navigation mark and transmit the monitoring status of each part to the ground central processor through wireless signals. When the ground central processor receives the signal sent by the navigation mark l, it is expressed as:

[0135]

[0136] The transmission rate from the navigation mark l to the central processing system is expressed as:

[0137] R l，u =B l，u log 2 (1 + γ l，u );

[0138] Among them, p l represents the signal transmission power of the navigation mark l, x l represents the transmitted signal of the navigation mark l, h l，u represents the channel gain between the navigation mark l and the ground station, where the noise is the power of the noise; represents the transmission signal-to-noise ratio, B l，u represents the transmission bandwidth, and the faster the transmission rate, the better the transmission performance of the system.

[0139] Example 2. As shown in the appendix Figure 2 This application also proposes a self-cleaning bird repelling system for navigation marks. The system includes:

[0140] A data acquisition module for collecting image data within a preset monitoring range and performing data annotation to obtain a target data set;

[0141] A first processing module that inputs the target data set into a pre-trained deep learning model to identify existing birds and calculates the corresponding bird information;

[0142] A second processing module that inputs the target data with existing birds into an image recognition model for bird droppings detection to obtain the droppings coverage rate;

[0143] A first control module that selects to activate the first bird repelling mode or the second bird repelling mode according to the calculated bird information; a second control module that selects to activate the first cleaning mode or the second cleaning mode according to the calculated droppings coverage rate;

[0144] And a backhaul module that uses orthogonal frequency division multiple access technology as the communication link between the ground central processing system and the navigation mark.

[0145] The system proposed in this application can accurately monitor the behavior of birds in an efficient, precise, and intelligent operation mode. It can automatically select appropriate bird repelling and cleaning strategies through advanced image recognition technology, deep learning models, automatic control, and efficient communication technology, maximizing work efficiency, reducing operating costs, and minimizing manual intervention.

[0146] Embodiment 3: This application also proposes a computer storage medium, in which a computer program is stored. The computer program is configured to execute the self-cleaning bird repelling method of a navigation mark as described above when running.

[0147] The computer program in the storage medium can automatically implement the self-cleaning bird repelling method of the navigation mark when running by executing the functions of the above-mentioned modules. This method combines image recognition, deep learning, intelligent control, and communication technology, enabling the system to execute tasks efficiently and intelligently without manual intervention, thereby improving the maintenance efficiency and automation level of the navigation mark.

[0148] Although example embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above example embodiments are merely exemplary and are not intended to limit the scope of this application. Those of ordinary skill in the art can make various changes and modifications therein without departing from the scope and spirit of this application. All such changes and modifications are intended to be included within the scope of this application as claimed in the appended claims.

[0149] It should be noted that in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0150] Although the description of this application is made in conjunction with the above specific embodiments, it is obvious that those skilled in the art can make many substitutions, modifications, and variations based on the above content. Therefore, all such substitutions, modifications, and variations are included within the spirit and scope of the appended claims.

Claims

1. A self-cleaning bird-repelling method for a navigation mark, characterized in that: The method comprises: S1: collecting image data of a preset monitoring range, and performing data annotation to obtain a target data set; wherein the target data set includes information of different types of birds and environmental information; S2: Inputting the target data set into a pre-trained deep learning model to identify the existing birds, and calculating and obtaining the corresponding bird information, wherein the bird information includes the bird species and location; S3: Input the target data of bird existence into the image recognition model to detect bird droppings and obtain the droppings coverage rate; S4: According to the bird information calculated by S2, the first bird repelling mode or the second bird repelling mode is selected to be started; at the same time, according to the feces coverage rate calculated by S3, different cleaning modes are selected.

2. A self-cleaning bird-repelling method for a navigation mark according to claim 1, characterized in that: The step S1 of performing data labeling also includes: The collected image data are annotated using a public data set; wherein the public data set is at least a Birdsnap data set or a CUB-200 data set; and the data annotation includes existing bird annotations and possible bird annotations.

3. A self-cleaning bird-repelling method for a navigation mark according to claim 2, characterized in that: Before step S2, the following steps are also included: Preprocessing the collected image data, the preprocessing including removing noise from the target data set and adjusting brightness and contrast; In the image data obtained after preprocessing, bird droppings are grayish white; navigation marks are black, red or yellow.

4. A self-cleaning bird-repelling method for a navigation mark according to claim 3, characterized in that: The step S3 further includes: S31: Convert the image from BGR to HSV or Lab color space; S32: according to the color characteristics of the bird droppings, setting a threshold to extract the droppings area; S33: After extracting the feces area, dilation and erosion operations are used to remove noise and fill small holes, findContours function is used to detect the extracted guano area, and Gaussian blur method is used for smoothing and denoising to optimize the segmentation effect.

5. A self-cleaning bird-repelling method for a navigation mark according to claim 4, characterized in that: The step S3 further includes: after segmenting the area covered by feces, calculating the proportion of feces in the navigation mark area in the entire image, the formula is: Feces coverage = (number of pixels in the feces area / number of pixels in the entire base area) × 100%; According to the feces coverage rate, the cleaning machine is divided into A gear, B gear and C gear, and corresponding cleaning modes are set according to different gears; Among them, A level has a low coverage rate and the cleaning work is completed; B level has a high coverage rate and cleaning needs to be started; C level has a very high coverage rate and requires manual cleaning.

6. A self-cleaning bird-repelling method for a navigation mark according to claim 5, characterized in that: The step S4 selects to start the first bird-repelling mode or the second bird-repelling mode according to the bird information calculated in S2, and further includes: When the bird is above or around the navigation mark, the first cleaning mode is started; the first cleaning mode is to turn on the variable frequency ultrasonic device, and the ultrasonic power amplifier frequency continuously changes from 20Khz to 70Khz; When the bird is located on the base below the navigation mark, the second cleaning mode is activated, wherein the second cleaning mode triggers the high-pressure water flow and air flow device so that the impact range of the water flow and air flow can cover the entire navigation mark base.

7. A self-cleaning bird-repelling method for a navigation mark according to claim 6, characterized in that: In step S4, different cleaning modes are selected according to the feces coverage rate calculated in S3, and the method further includes: dividing the overall operation time into T working time slots during the cleaning process of the system, each unit t time slot is a cleaning cycle, and each beacon is used as an intelligent agent to interact with the environment to make adaptive cleaning action decisions; the beacon intelligent agent will make the system transition to a new state by performing different cleaning actions, and in the next time slot, the intelligent agent will make a better cleaning action by analyzing the new state, and finally, through continuous interaction with the environment, the intelligent agent will obtain a stable reward, and at this moment, the beacon intelligent agent learns to execute adaptive cleaning actions to make the best cleaning action in each time slot; specifically including: The state space, action space and reward are defined as follows: According to the bird-repelling cleaning process, the state space is defined as the feces coverage C, the number of cleaning operations n l ;Then the state of the tth time slot is expressed as s(t)=[C(t),N(t)]; Where C(t)=[c1(t),...,c l (t),...,c L (t)], represents the feces coverage matrix of the navigation mark at time slot t; N(t) = [n1(t),...,n l (t),...,n L (t)] represents the number of cleaning times of the beacon in each time slot; The action of each time slot is defined as the scheduling strategy for beacon cleaning, expressed as a(t) = [K(t), F(t)]; where K(t)=[k1(t),...,k l (t),...,k L (t)], represents the cleaning scheduling execution matrix of each time slot beacon node, where k l (t) = {0, 1, 2} represents the scheduling status of the cleaning function in each time slot, 0 means no cleaning is needed at this moment, 1 means cleaning needs to be started at this time slot, and 2 means manual cleaning is required; F(t) = [f1(t),...,f1(t),...f L (t)], indicating the number of times each beacon needs to be sprayed with water for each cleaning, where f1(t) = {0, 1, 2, 3}; The reward at time slot t is defined as a linear function of the overall utility of the system observed at the end of time slot t, expressed as r(t) = -λΣ l∈L c l (t), where λ is a constant, expressed as the inverse of the area covered by bird droppings on the buoy; so the sum of future returns can be defined as: Among them, T and 0<γ<1 are the last step of each cycle and the discount factor representing the impact of future rewards respectively; The DQN framework is used to optimize the strategy, where DQN uses an artificial neural network Q(s,a;θ), called a prediction network, as a function approximator to estimate the action value function, and θ is the weight of the neural network; The input of the prediction network is the state s, and the corresponding values ​​of all possible actions are generated as output; The target network is a neural network Q'(s,a;θ'), which is used to estimate the target value; the target network has the same structure as the prediction network; the weight θ' of the target network is copied from θ at every fixed number of iterations n instead of every training cycle; Depending on the actions taken by the beacon agent, the beacon cleaning environment will transition to a new state s(t+1); the agent collects immediate rewards r(t) from the environment and updates the weights of the neural network in the algorithm by minimizing the loss function L(θ) at each step: L(θ)=[(y j -Q(s j ,a j ;i)) 2 ];y j =r j +γmax a' Q'(s j+1 ,a';θ'); Eventually, the neural network will converge to a state where the reward tends to be stable. At this time, the beacon agent will learn to adaptively adjust its cleaning actions to adapt to the dynamic cleaning environment.

8. A self-cleaning bird-repelling method for a navigation mark according to any one of claims 1 to 7, characterized in that: The method further includes: using orthogonal frequency division multiple access technology as an access scheme for different communication links; specifically: Define a set of lake navigation marks as Where l represents the lth beacon; each beacon has a terminal device monitoring function, which can monitor the various system functions on the beacon and transmit the monitoring status of each part to the ground central processor through wireless signals. The ground central processor receives the signal sent by beacon l, which is expressed as: The transmission rate from beacon l to the central processing system is expressed as: R l，u =B l，u log2(1+γ l，u ); Among them, p l represents the signal transmission power of beacon l, x l Indicates the signal sent by the beacon l, h l，u represents the channel gain between the beacon l and the ground station, where the noise is the power of the noise; represents the transmission signal-to-noise ratio, B l，u Represents the transmission bandwidth. The faster the transmission rate, the better the transmission performance of the system.

9. A self-cleaning bird-repelling system for a self-cleaning bird-repelling method for a navigation mark as claimed in any one of claims 1 to 8, characterized in that: The system comprises: The data acquisition module is used to collect image data within the preset monitoring range, and to perform data annotation to obtain the target data set; The first processing module inputs the target data set into a pre-trained deep learning model to identify the existing birds and calculate the corresponding bird information; The second processing module inputs the target data of bird existence into the image recognition model to detect bird droppings and obtain the droppings coverage rate; The first control module selects to start the first bird-repelling mode or the second bird-repelling mode according to the calculated bird information; the second control module selects to start the first cleaning mode or the second cleaning mode according to the calculated feces coverage rate; As well as the backhaul module, it uses orthogonal frequency division multiple access technology as the communication link between the ground central processing system and the navigation mark.

10. A computer storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute a self-cleaning and bird-repelling method for a navigation mark as described in any one of claims 1 to 8 when running.

Images